US-12627837-B2 - Implementation efficient partition-based intra coding concept

US12627837B2US 12627837 B2US12627837 B2US 12627837B2US-12627837-B2

Abstract

A method for decoding a block of a picture includes decoding, from a data stream, an intra-coding mode for the block, deriving a predictor for a prediction partition of the block using at least one already reconstructed samples neighboring the prediction partition and the intra-coding mode, deriving a prediction residual for each of the at least two transform partitions included in the prediction partition, and combining the predictor of the prediction partition and the prediction residual for each of the at least two transform partitions included in the prediction partition to reconstruct the prediction partition. The block is partitioned into multiple transform partitions based on a partition dimension flag, and the prediction partition includes at least two transform partitions of the multiple transform partitions.

Inventors

Santiago DE LUXÁN HERNÁNDEZ
Benjamin BROSS
Phan Hoang Tung NGUYEN
Valeri GEORGE
Heiko Schwarz
Detlev Marpe
Thomas Wiegand

Assignees

Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V.

Dates

Publication Date: 20260512
Application Date: 20240308
Priority Date: 20190312

Claims (20)

1 . A method for decoding a block of a picture, the method comprising: decoding, from a data stream, an intra-coding mode for the block, wherein the block is partitioned into four equally sized transform partitions based on a partition dimension flag and a size of the block; determining, based on the size of the block, a plurality of prediction partitions that the block is divided; deriving a predictor for a first prediction partition of the plurality of prediction partitions using at least one already reconstructed sample neighboring the first prediction partition and the intra-coding mode, wherein the first prediction partition corresponds to two of the transform partitions; after deriving the predictor for the first prediction partition, deriving a first prediction residual for a first transform partition and a second prediction residual for a second transform partition, wherein the first and the second transform partitions correspond to the first prediction partition; and combining the predictor of the first prediction partition with the first prediction residual and the second prediction residual to reconstruct the first prediction partition.
2 . The method of claim 1 , further comprising: reconstructing the block using the reconstructed prediction partition, a prediction residual for each of the four transform partitions, and a predictor of each additional prediction partition of the plurality of prediction partitions.
3 . The method of claim 1 , further comprising: decoding, from the data stream, the partition dimension flag indicating a partitioning direction that the block is divided; and determining that the block is divided along the partitioning direction, indicated by the partition dimension flag, into the four equally sized transform partitions, wherein each of the transform partitions span a vertical height of the block when the partitioning direction is horizontal or span a horizontal width of the block when the partitioning direction is vertical.
4 . The method of claim 3 , wherein: a first dimension of each of the transform partitions and a first dimension of the block are the same while a second dimension of the transform partitions and a second dimension of the block are different.
5 . The method of claim 3 , wherein: when the size of the block is 8×4 or 4×8, the method comprises determining that the block is divided into two equally sized transform partitions.
6 . The method of claim 1 , wherein: the method further comprises decoding, from the data stream, the partition dimension flag indicating a partitioning direction that the block is divided; when the partition dimension flag indicates that the partitioning direction of the block is horizontal, the method comprises determining that a size of each of the transform partitions is defined as ( W × H K ) ; and when the partition dimension flag indicates that the partitioning direction of the block is vertical, the method comprises determining that a size of each of the transform partitions is defined as ( W K × H ) , where W is a width of the block, H is a height of the block, and K is four.
7 . The method of claim 1 , wherein: the intra-coding mode is one out of a set of supported intra-coding modes including angular modes and at least one non-angular mode; the first prediction partition includes at least sixteen samples; the first transform partition is a first size; and the first prediction partition is a second size that is different than the first size.
8 . The method of claim 1 , wherein deriving the first prediction residual and the second prediction residual, comprises: decoding a transform of the prediction residual from the data stream; or inferring a value of the prediction residual.
9 . The method of claim 1 , wherein: the plurality of prediction partitions is two.
10 . The method of claim 1 , further comprising: deriving a predictor for each additional prediction partition of the plurality of prediction partitions using at least one already reconstructed sample and the intra-coding mode.
11 . The method of claim 10 , wherein the at least one already reconstructed sample used for predicting each additional prediction partition are: external to and neighboring the block, or external to and neighbor each of the additional prediction partitions, respectively.
12 . The method of claim 1 , wherein: reconstructed samples at a location corresponding to one of the four transform partitions included in the reconstructed prediction partition are not used to derive a predictor corresponding to another prediction partition of the plurality of prediction partitions; and reconstructed samples at a location corresponding to another one of the four transform partitions included in the reconstructed prediction partition are used to derive a predictor corresponding to another prediction partition of the plurality of prediction partitions.
13 . The method of claim 1 , wherein a prediction residual for each of the four transform partitions is derived after the predictor for the prediction partition is derived.
14 . A decoder for decoding a block of a picture, the decoder comprising a processor configured to: decode, from a data stream, an intra-coding mode for the block, wherein the block is partitioned into four equally sized transform partitions based on a partition dimension flag and a size of the block; determine, based on the size of the block, a plurality of prediction partitions that the block is divided; derive a predictor for a first prediction partition of the plurality of prediction partitions using at least one already reconstructed sample neighboring the first prediction partition and the intra-coding mode, wherein the first prediction partition corresponds to two of the transform partitions; after deriving the first predictor for the prediction partition, derive a first prediction residual for a first transform partition and a second prediction residual for a second transform partition, wherein the first and the second transform partitions correspond to the first prediction partition; and combine the predictor of the first prediction partition with the first prediction residual and the second prediction residual to reconstruct the first prediction partition.
15 . The decoder of claim 14 , wherein the decoder is further configured to: reconstruct the block using the reconstructed prediction partition, a prediction residual for each additional transform partition of the four transform partitions, and a predictor of each additional prediction partition of the plurality of prediction partitions.
16 . The decoder of claim 14 , wherein the decoder is further configured to: decode, from the data stream, the partition dimension flag indicating a partitioning direction that the block is divided; and determine that the block is divided along the partitioning direction, indicated by the partition dimension flag, into four equally sized transform partitions, wherein each of the transform partitions span a vertical height of the block when the partitioning direction is horizontal or span a horizontal width of the block when the partitioning direction is vertical.
17 . The decoder of claim 16 , wherein a first dimension of each of the transform partitions and a first dimension of the block are the same while a second dimension of the transform partitions and a second dimension of the block are different.
18 . The decoder of claim 16 , wherein: when the size of the block is 8×4 or 4×8, the decoder is configured to determine that the block is divided into is two equally sized transform paritions.
19 . The decoder of claim 14 , wherein: the decoder is further configured to decode, from the data stream, the partition dimension flag indicating a partitioning direction that the block is divided; when the partition dimension flag indicates that the partitioning direction of the block is horizontal, the decoder is configured to determine that a size of each of the transform partitions is defined as ( W × H K ) ; and when the partition dimension flag indicates that the partitioning direction of the block is vertical, the decoder is configured to determine that a size of each of the transform prediction partitions is defined as ( W K × H ) , where W is a width of the block, H is a height of the block, and K is four.
20 . The decoder of claim 14 , wherein: the intra-coding mode is one out of a set of supported intra-coding modes including angular modes and at least one non-angular mode; the first prediction partition includes at least sixteen samples; the first transform partition is a first size; and the first prediction partition is a second size that is different than the first size.

Description

CROSS-REFERENCES TO RELATED APPLICATIONS This application is a continuation of U.S. patent application Ser. No. 17/471,911, filed Sep. 10, 2021, which in turn is a continuation of International Patent Application No. PCT/EP2020/056489, filed Mar. 11, 2020, each of which is incorporated herein by reference in its entirety, and additionally claims priority from European Patent Application No. 19162405.5, filed Mar. 12, 2019, which is incorporated herein by reference in its entirety. The present application is concerned with an intra-coding concept for use in a block-based codec such as, for example, a hybrid video codec. BACKGROUND OF THE INVENTION Given a certain block, intra prediction is carried out in HEVC by extrapolating the decoded boundary samples of the neighboring blocks following certain patterns, namely 33 angular modes and a DC and a planar modes [1]. The one intra prediction mode that minimizes the rate-distortion cost is then signaled to the decoder. Despite the known codecs supporting many Intra Prediction Modes (IPMs), the intra prediction achieved thereby is still subject of development to find better intra predictors leading to higher coding efficiency. This does not only pertain to HEVC but also to other block-based codecs using intra-prediction. Finding a set of intra-prediction modes which are suitable for efficiently coding the inner of blocks requires taking into account the overhead for signaling the intra-prediction mode in terms of signaling overhead and the resulting quality of the predictor obtained by these intra-prediction modes due to the fact that a more accurate predictor reduces the prediction residual, thereby reducing the signaling overhead associated with coding the prediction residual. In order to keep the signaling overhead associated with the intra-prediction modes low, intra-predicted blocks should be large, i.e. the granularity at which the intra prediction mode is signaled should be kept coarse, but on the other hand, spatial prediction of larger blocks tends to be less accurate owing to a higher mean sample distance of the samples in the inner of the intra-predicted block, i.e. the ones to be predicted, to the already decoded/encoded samples neighboring this block, i.e. the reference samples. HEVC alleviates this catch-22 a little bit by allowing the transform residual blocks to inherit the intra-prediction mode of their corresponding coding unit relative to which the transform residual blocks form leaf blocks into which the coding unit is sub-divided by multi-tree subdivisioning. However, this still requires signaling overhead for signaling from encoder to decoder the sub-partitioning of respective intra-coded coding units into the transform blocks. A newly developing intra coding concept is presented by the Intra Sub-Partitions (ISP) coding mode in the newly developing Versatile Video Coding (VVC) standard, but here, implementation efficiency improvements would be required. SUMMARY An embodiment may have a decoder for block-based decoding of a picture from a data stream, configured to decode an intra-coding mode for a predetermined block of the picture from the data stream; decode a partition dimension flag for the predetermined block of the picture from the data stream and set a partition dimension depending on the partition dimension flag to be horizontal or vertical, partition, along the predetermined dimension, the predetermined block into transform partitions which are as wide as the predetermined block perpendicular to predetermined dimension; decode, for each transform partition, a transform of a prediction residual from the data stream; intra-predicting the predetermined block depending on one or more already reconstructed samples neighboring the predetermined block in a manner depending on the intra-coding mode to acquire a predictor for the predetermined block; and reconstructing the predetermined block by correcting the predictor within each transform partition using the transform of the prediction residual decoded for the respective transform partition. Another embodiment may have an encoder for block-based encoding of a picture into a data stream, configured to encode an intra-coding mode for a predetermined block of the picture into the data stream; encode a partition dimension flag for the predetermined block of the picture into the data stream which signals that a partition dimension is to be set to be horizontal or vertical, partition, along the predetermined dimension, the predetermined block into transform partitions which are as wide as the predetermined block perpendicular to predetermined dimension; intra-predicting the predetermined block depending on one or more already reconstructed samples neighboring the predetermined block in a manner depending on the intra-coding mode to acquire a predictor for the predetermined block; encode, for each transform partition, a transform of a prediction residual into the data stream, so that the predeter